Higher risk, lower returns: What hedge fund investors really earn

Table 2 provides evidence on buy-and-hold vs. dollar-weighted returns at the level of the individual fund. Panel A provides results for the combined sample that includes all 10,954 hedge funds and hedge fund-like entities. There are two highlights in Panel A. First, mean buy-and-hold return across funds is 6.1 percent while the mean dollar-weighted return is only 2.9 percent, implying a statistically significant and economically substantial 3.2 percent performance gap between buy-and-hold measures and actual investor returns.⁸ Thus, the results in Panel A confirm earlier impressions that actual hedge fund investor returns are considerably lower than existing estimates based on geometric averages. Note that the absolute magnitude of both returns is low, especially compared with the descriptive statistics in Table 1. The poor returns in Table 2 are due to the inclusion of a great multitude of small, short-lived and poor-performing funds, especially in the most recent years of the sample.

Panel A also presents the dispersion of returns across funds. Note that providing an estimate of return variability is challenging for dollar-weighted specifications because dollar-weighting is essentially a time-series phenomenon, and one needs at least 10 to 15-year investment horizons to provide a meaningful distinction from buy-and-hold returns. Thus, given the relatively short history of hedge fund investing, there is simply not enough information to estimate meaningful variability in the time-series of dollar-weighted returns. Instead, we use the cross-section of funds to provide an estimate of cross-fund variability of dollar-weighted returns. This specification also has a natural real-world investment interpretation: essentially, it captures the risk that investors face by choosing one fund versus another to invest in. We examine the dispersion of cross-fund returns on two dimensions: first, we look at the standard deviation of returns; second, since hedge fund returns may be not well-behaved, we examine the properties of the full empirical distribution of returns.

The second major finding in Panel A of Table 2 is that the dispersion of dollar-weighted returns is higher than that of buy-and-hold returns. Specifically, buy-and-hold returns have a standard deviation of 18.7 percent across funds, while the standard deviation of dollar-weighted returns is 20.2 percent; the resulting difference of 1.5 percent is statistically significant but seems economically modest. An examination of the percentiles of the empirical distributions of the two return metrics in Panel A confirms impressions from summary mean and standard deviation statistics, and suggests that outliers and other distributional quirks cannot account for the observed results. For example, tests on median returns confirm a reliable difference between buy-and-hold and dollar-weighted returns. Note also that the two distributions of returns almost match each other in their upper percentiles. However, there is a growing gap in the lower percentiles of the returns distributions, which reaches over 10 percent at the first percentile. Thus, the bottom line from Panel A is that actual investor returns are substantially lower and somewhat more variable than estimates based on the buy-and-hold assumption.⁹ These results suggest that hedge fund investors take higher risks and earn lower returns than previously thought.

As mentioned above, a disadvantage of the results in Panel A is that all funds are weighted equally, while there are great differences in fund capitalization and longevity, with corresponding differences in fund importance to investors. To provide a more apples-to-apples distribution of returns across funds, we investigate two subsamples of funds in Panel B, comprising funds with at least a 5-year and 10-year record, respectively. These requirements result in substantially reduced sample sizes but are also more representative of the return experience of the “typical” investor. Since the results are largely the same across these two subsamples, we only discuss the 5-year specification. We find that buy-and-hold and dollar-weighted returns are both higher for this specification but the performance wedge remains almost the same at 3.1 percent. The standard deviation of dollar-weighted returns is again higher than that of buy-and-hold returns and now the difference looks more material both on an absolute basis (2.2 percent), and especially as compared to the base variability of these larger, more stable funds (about 10 to 12 percent). Results based on the percentiles of the empirical distribution of returns are consistent with those for the mean and the standard deviation; for parsimony, we omit them for the rest of the table.

Next, we examine the mean returns of individual funds in different sub-periods to examine the robustness of our findings to time-series factors. Specifically, we examine returns over early (1980 to 1994) vs. later years (1995 to 2008), and also the effect of excluding the pivotal year 2008. Panel C shows that dollar-weighted returns are lower and more variable than buy-and-hold returns in all of these specifications. As expected, both return estimates are much higher after excluding year 2008 but the dollar-weighted wedge remains largely the same.

There is considerable heterogeneity in the universe of hedge fund vehicles, including hedge funds proper, funds-of-funds, commodity pool operators (CPOs) and commodity trading advisors (CTAs). Panel D of Table 2 provides a breakdown of results based on these categories. There is some variation in the relative magnitude of the results across categories but the same basic pattern is largely confirmed. In addition, in untabulated results we find largely the same dollar-weighted effects in a split between Active and Inactive funds. Summarizing, the results for individual fund show that dollar-weighted returns are reliably lower than buy-and-hold returns on the magnitude of 3 percent. Dollar-weighted returns also tend to be more variable but this effect is economically modest.
3.2 Portfolio-level dollar-weighted returns

A shortcoming of the results in Table 2 is the equal weighting allotted to each fund regardless of length of existence or amount of capital employed. Table 3 addresses this shortcoming using value-weighted portfolio specifications, where buy-and-hold returns are computed as the geometric average of the individual years’ value-weighted returns over all available funds. Thus, the buy-and-hold calculation in Table 3 is value-weighted in the cross-section of the available fund sample (but equally-weighted in the time-series of returns). Dollar-weighted returns are computed by aggregating the individual funds’ capital flows, and computing an IRR over the initial aggregate assets-under-management, the monthly aggregate capital flows, and the ending aggregate assets of the portfolio of available funds. Thus, the dollar-weighted calculation is value-weighted in both the cross-section and the time-series of returns. Because the results in Table 3 properly reflect the longevity of the funds and the effect of amount of invested capital, we view them as most representative of the average investor experience and therefore as the main results of the paper.

Panel A in Table 3 exhibits the aggregate results for all funds. While the buy-and-hold return is a solid 12.6 percent, the dollar-weighted return is only 6.0 percent, for a very substantial performance gap of 6.6 percent. We assess the statistical significance of this difference using a bootstrap test. The advantage of bootstrap tests is that they avoid the usual distributional assumptions, which is especially relevant given the properties of hedge fund returns. The test is based on deriving a 1,000-observation bootstrap distribution of the test statistic under the null that fund flows do not matter for the calculation of dollar-weighted returns, and comparing the observed difference between buy-and-hold and dollar-weighted returns to the derived bootstrap distribution, see Appendix B for an expanded description and explanation. The p-value of this test is 1.2 percent for the aggregate portfolios in Panel A, revealing reliable statistical significance.

Similar to the preceding analyses at the individual fund level, we present results for several subperiods. Buy-and-hold returns are much higher during 1980-1994 than during 1995-2008 but the dollar-weighted wedge is material in both subperiods (4.8 and 2.9 percent respectively). Note that the dollar-weighted wedge within the two subperiods is considerably lower than the one over the whole sample. This happens because dollar-weighting is essentially a time-series phenomenon, and thus restricting the time-series almost by definition restricts the dollar-weighted wedge as well. The material difference between the magnitude of the dollar-weighted wedge for the whole sample and within subperiods indicates that there are material dollar-weighted effects across the 1980-1994 and 1995-2008 subperiods.

As expected, the magnitude of returns is higher when excluding the pivotal year 2008, and the dollar-weighted wedge shrinks from 6.6 percent to 4 percent. This evidence suggests that the dramatic events of 2008 had a much worse effect on investors than that suggested by traditional metrics; this is to be expected given that investors’ capital exposure peaked in 2007 (see Table 1), exactly the worst time to be heavily invested in hedge funds. Thus, the experience of year 2008 is a vivid illustration of the importance of dollar-weighting. Using buy-and-hold metrics, the 2008 experience looks unpleasant but only mildly so, with average returns declining from 13.8 percent as of the end of 2007 to 12.6 percent as of the end of 2008. Dollar-weighting, which properly reflects the peak capital exposure of investors as of 2007, paints a much bleaker picture, with average returns declining from a respectable 9.7 percent to a disappointing 6 percent, not that different from risk-free rates over the 1980-2008 period.

As discussed above, it is well-known that hedge fund data suffers from serious self-selection biases because hedge funds self-report their performance, where specific examples include incubation bias and backfill bias (Brown, Goetzmann, and Ibbotson 1999 and Teo 2009). Incubation bias arises because hedge funds rely mostly on internal capital during their early years, and later successful funds attract much outside capital and publicize their returns while we do not observe the returns of the unsuccessful funds. The related backfill bias arises when database providers backfill the returns of newly entering funds, resulting in an inflated estimate of realized returns. It is less clear, however, whether these biases affect just the absolute level of returns or the dollar-weighted wedge as well. It is possible that since investors chase past returns of outperforming funds, backfill bias can explain some or even most of the difference in dollar-weighted and buy-and-hold returns.

We address the incubation and backfill bias in two ways, one quite stringent but perhaps too restrictive, and the second one taking a more moderate path. For the stringent one, we retain only observations for which we are sure there is no backfill problem; specifically, Lipper-TASS provides a start-of-reporting date and we eliminate all observations before that, while CISDM does not provide such a date, so we eliminate all CISDM observations. The buy-and-hold return for the stringent specification is 11.7 percent in Panel A of Table 3, which is slightly lower than the buy-and-hold return for the whole sample (12.6 percent), consistent with backfill bias inflating performance. For our purposes, the key is that the 5.0 percent difference between buy-and-hold and dollar-weighted returns remains statistically and economically significant. For the more moderate specification, we follow Teo (2009) and drop the first twelve months of available fund returns; the tenor of the results remains unchanged. Thus, incubation and backfill bias seem to have only a minor effect on the calculated dollar-weighted effects.

Panel B in Table 3 breaks down the value-weighted results of Panel A by type of fund. Hedge funds proper are the largest group and also have the highest buy-and-hold returns at 13.8 percent, while fund-of funds have the lowest corresponding return at 11.0 percent. The pattern found for aggregate returns in Panel A is confirmed for the partitions in Panel B, where all subgroups have dollar-weighted returns lower than buy-and-hold returns, and this performance gap is on the magnitude of 4 to 8 percent, highly significant in bootstrap tests except for the limited sample of CTAs and CPOs.

A chronic difficulty in evaluating hedge fund returns is finding appropriate benchmarks. Hedge funds comprise a number of disparate and sometimes exotic assets classes and strategies, including investing in stocks, real estate, and venture capital, and using options, substantial leverage, and short positions; thus, it is challenging to properly assess their risk profile and the commensurate return. To some extent, dollar-weighted returns are themselves a natural solution to benchmarking problems because there is no better control for a fund’s risk profile than the fund itself. Accordingly, most of the analyses in this study emphasize the comparison between the fund buy-and-hold and dollar-weighted returns, which properly and fully reflects the difference between investment and investor returns.

Hedge fund investments, however, are an organic and interchangeable part of the larger world of possible investments, and thus some comparison with external benchmarks is warranted. We accomplish this task on two dimensions. First, in Panel A of Table 4 we present a simple comparison of aggregate portfolio dollar-weighted returns (as in Table 3) with returns on the S&P 500 index and the risk-free rate (measured as the 1-month T-bill rate). We also include a hypothetical dollar-weighted return using hedge funds’ pattern of capital flows combined with the return of the S&P 500; the motivation is to provide an “investment alternatives” benchmark for what hedge fund investors would have earned if they had invested in the S&P 500 instead. Given the dramatic effect of year 2008, we present results both including and excluding that year. An examination of the results in Panel A reveals that hedge fund dollar-weighted returns are substantially lower than the returns on the S&P 500 and only marginally higher than the risk-free rate of return over 1980-2008. Dollar-weighted returns look better excluding year 2008 but are still reliably within the spread of the risk-free rate and S&P 500 return. Note that the hypothetical return calculated with hedge fund flows and the S&P 500 returns is by far the lowest in Panel A. This result confirms earlier impressions that it is not so much the investment but poor capital flow timing which causes the low returns; specifically, poor timing is what causes hedge fund investors to earn lower returns than the funds, and this same timing would have brought them poor returns on the broad stock market as well.

A disadvantage of the analysis in Panel A is that the benchmarks considered there are only a crude reflection of the investment profile of hedge funds. Existing research has developed more sophisticated models of hedge fund benchmarks, and thus better estimates of hedge fund alpha after controlling for exposure to various (risk) factors, e.g., Agarwal and Naik (2004), Edwards and Caglayan (2001) and Fung and Hsieh (2004). Thus, for our second approach we compare hedge fund alpha to the dollar-weighted wedge documented in this study; the intuition is that investors’ risk-adjusted return (or net alpha) is really fund alpha minus the dollar-weighted wedge. As discussed above, most existing evidence points to estimates of hedge fund alpha on the magnitude of 3 to 5 percent, e.g., Ibbotson and Chen (2006), Kosowski, Naik, and Teo (2007), Brown, Goetzmann and Ibbotson (1999). Using our estimate of the dollar-weighted wedge between 3 and 7 percent suggests that net alpha is close to zero or even negative.

To provide a more careful evaluation of this approach in our sample, in Panel B of Table 4 we use the Fama-French 3-factor model and the Fung and Hsieh (2004) 7-factor model to estimate hedge fund alphas and the resulting net alphas; the two models seem good complements for our study because the Fama-French model is more generic but available for longer periods, while the Fung-Hsieh model is more comprehensive and specifically developed for hedge funds but because of more stringent data requirements is available only for the second part of our sample period (1996-2008). We derive risk-adjusted returns by using a time-series regression, where the regression is estimated every month using the past 24 monthly returns to allow factor loadings to vary over time (see Appendix C for details on the estimation for the two models). The regression is run at the individual fund level and also at the value-weighted portfolio level, corresponding to our two main dollar-weighted specifications, at the fund level in Table 2 and at the portfolio level in Table 3. The results in Panel B reveal mean alphas of 1 to 4 percent across funds. Since the corresponding dollar-weighted wedge is on the magnitude of 2 to 4 percent in Table 2, the combined impression form these results is that net investor alpha is likely close to zero. Portfolio alphas in Panel B at 5 to 6 percent are substantially higher than means over funds, consistent with the value-weighted portfolio specification discounting the poor returns of short-lived, smaller funds. The corresponding value-weighted wedge in Table 3 is also higher, though, on the magnitude of 3 to 7 percent. Again, the resulting impression is that after accounting for dollar-weighted effects investors’ net alpha is likely negligible.

In this section, we explore the magnitude of dollar-weighted effects as a function of a number of salient fund characteristics. The goal is to check the robustness of the results and to identify possible environments and fund features where dollar-weighted effects are especially pronounced. For parsimony, we only present the results for hedge funds proper; results for other types of funds and all funds are generally similar. We rely on quintile specifications to map out the potential relations for two reasons. First, the quintile assignments based on ex ante variables have a natural investment portfolio interpretation and provide an immediate feel for the economic magnitude of the results; second, the portfolio specification allows us to identify possible non-linearities in the identified relations.

Panel A of Table 5 presents return results by fund size, where funds are ranked each year on prior year’s average market value into quintiles. The results in Panel A of Table 5 reveal a clear negative relation between size and buy-and-hold returns, probably because small funds are more able to invest in smaller and overlooked investment opportunities. This pattern is mirrored in the dollar-weighted specification, where the corresponding dollar-weighted returns are reliably and fairly uniformly lower than their buy-and-hold counterparts. The dollar-weighted wedge is a reliable feature across all fund sizes but does not vary much across quintiles.

In Panel B we present returns by volatility of buy-and-hold returns and volatility of capital flows. The motivation is that differences between buy-and-hold and dollar-weighted returns are likely to be more pronounced when there is more room for capital flow timing, i.e., when these two variables are larger in absolute magnitude. For both of these variables, the quintiles assignment is based on the last two years of available data, with a two-year minimum requirement of data availability. This additional data requirement leads to loss of some observations but seems necessary given that the distributions of returns and capital flows for hedge funds are possibly not well-behaved, so longer horizon of estimation is preferable. An inspection of Panel B reveals little in the way of reliable patterns for the returns and dollar-weighted differences across volatility of capital flows. For the volatility of buy-and-hold returns, though, there is a non-monotonic but fairly clear pattern of increased dollar-weighted wedge across quintiles, where the wedge for quintile 5 (9.9 percent) is almost double that for quintile 1 (5.2 percent). This result confirms the intuition that dollar-weighted effects are likely to be larger in environments with increased potential for (bad) timing of capital flows. In untabulated results, we also explored specifications based on the signed level of past 2-years’ buy-and-hold returns and capital flows; we found that the dollar-weighted difference persists in all specifications but no reliable patterns in its magnitude across quintiles.

Table 6 presents quintile results by level of fund fees and various contractual provisions restricting investor capital flows - variables which are more discretionary and are really part of fund design and positioning. In Panel A, we find some evidence that funds with higher fees earn moderately higher buy-and-hold returns, consistent with a functioning market for manager talent, where the rewards of superior performance flow to both fund managers and investors (recall that all return figures in the paper are net of fees). This pattern of superior performance for funds with higher fees is preserved in dollar-weighted returns, while the dollar-weighted wedge remains substantial but largely the same across quintiles. Panel B presents results by allowable frequency of redemption (annual, quarterly, monthly) and presence of lock-up period. There is some evidence of superior buy-and-hold performance for funds with the most stringent restrictions on redemptions, consistent with arguments that frequent investor redemptions can be distracting and counter-productive. Interestingly, this pattern is preserved and even magnified in dollar-weighted returns, with the lowest dollar-weighted wedge for funds with the most stringent restrictions. Thus, there is some evidence that limiting capital outflows is in the investors’ best interests as well.

Summarizing, we find statistically significant and economically substantial dollar-weighted differences for nearly all subsamples explored, which implies that dollar-weighted effects are a pervasive feature of hedge fund data. We find more limited evidence of differential dollar-weighted effects, with funds with high volatility of returns and liberal redemption policies the most prone to poor investor timing.

In this section, we examine more closely the nature and causes of dollar-weighted effects. First, we probe into the origins of the dollar-weighting effect by decomposing the fund-level performance gap into two drivers.¹⁰ As discussed earlier, the dollar-weighed wedge captures the aggregate effect of the hedge fund industry receiving continual infusions of capital while aggregate returns of hedge funds have been going down; this is an aggregate time-series effect. The dollar-weighted wedge also arises because hedge fund investors chase past performance across individual funds; this is a cross-sectional effect.¹¹ We disentangle the relative magnitude of the two effects by examining the aggregate time-series effect while holding the cross-sectional effect constant. This is accomplished by computing a hypothetical dollar-weighted return where monthly fund flows are assumed to be the same across all funds (as a percentage of beginning AUM), and are equal to the aggregate flow over the aggregate beginning AUM. For each fund, we recalculate the monthly capital flows under this assumption and compute the corresponding hypothetical dollar-weighted return. We measure the aggregate time-series effect as the mean difference between each fund’s buy-and-hold return and the hypothetical dollar-weighted return, while the remaining difference between the hypothetical and the actual dollar-weighted return captures the cross-sectional effect.

The results for this decomposition are presented in Table 7. Since the computations are at the fund level, we use the same fund-level sample as in Table 2 for clarity and continuity. Accordingly, the buy-and hold and dollar-weighted returns are the same as in Table 2, while the hypothetical dollar-weighted return and the decomposition of the dollar-weighted wedge are computed as explained above. We concentrate on the level-of-return results since the dollar-weighted effects for volatility are relatively modest. An examination of Panel A reveals that both the time –series and the cross-sectional effect play a role in explaining the total dollar-weighted effect. However, the aggregate time-series effect dominates the cross-sectional effect in our sample; the results differ across specifications but the time-series effect is always between about 50 to 75 percent of the total dollar-weighted effect.

Next, in Table 8 we provide evidence on the return-chasing vs. return-predicting role of capital flows in explaining the dollar-weighted performance gap. Recall that dollar-weighted returns are lower if beginning (discounted) asset holdings are negatively related to current period’s returns. This happens when the fund inflows of the current period are either positively related to past returns or negatively related to future returns; of course, the converse applies for capital outflows but the exposition emphasizes capital inflows for parsimony and because they dominate empirically. We explore the empirical magnitude of these past/future relations as explanations for the performance gap. Specifically, Table 8 presents the empirical distribution of the correlation of capital flows and past and future returns for all individual funds over 3-year horizons. The mean correlation of Capital flows/AUM and prior years’ return is reliably negative, steadily decreasing in absolute magnitude from -0.26 to -0.15 as horizons lengthen from t-1 to t-3. Having in mind that negative investor capital flows signify fund inflows, the documented negative correlation signifies that hedge fund investors chase returns. In contrast, the mean correlation of scaled capital flows and future years’ return is essentially zero. Medians and the rest of the empirical distributions show similar patterns for both past and future returns, suggesting that these results are robust.¹² Thus, the dollar-weighted performance gap seems predominantly driven by investors’ return-chasing behavior.